Data streams fusion by frequent correlations mining

Abstract

Applications acquiring data from multiple sensors have to properly refer data to observables. On-line classification and clustering as basic tools for performing information fusion are computationally viable. However, they poorly exploit temporal relationships in data as patterns mining methods can do. Hence, this paper introduces a new algorithm for the correlations mining in the proposed graph-stream data structure. It can iteratively find relationships in complex data, even if they are partially unsynchronized or disturbed. Retrieved patterns (traces) can be used directly to fuse multi-perspective observations. The algorithm’s evaluation was conducted during experiments on artificial data sets while its computational efficiency and results quality were measured.